Jefferies' Chris Wood: US AI Spending Most Likely Ends in 'Massive Capital Destruction'

Chris Wood, global head of equity strategy at Jefferies, put a number on his skepticism about the AI buildout in his GREED & fear newsletter dated Friday, Oct. 9. His view is that the most likely long-term outcome is "massive capital destruction" in the US, with market share shifting to cheaper open-source Chinese models.

Wood did not say demand for AI is fading. His argument is about price and returns.

The spending

US hyperscalers have issued combined capital expenditure guidance of about $695 billion for 2026. That is projected to climb to $870 billion in 2027.

Hyperscalers are the giant cloud operators that rent out computing power. Capex is what they spend on data centers, chips and related hardware.

Wood calls much of that outlay malinvestment, meaning money put into projects unlikely to earn back its cost.

The Chinese competition

"Initially the focus was on the cheapness of China models and the related commoditisation threat facing large language models," Wood wrote. "But now there is also a growing realisation that China has become a technological peer to the US in AI."

He points to usage figures. Chinese models processed 36.39 trillion tokens in the week ending July 19, 2026. Top US models handled 7.39 trillion over the same week. Tokens are the chunks of text a model reads and generates, a rough gauge of usage.

Price is part of the gap. Moonshot AI's Kimi K3 and Zhipu AI's GLM-5.2 cost roughly one-quarter per token of some US alternatives, according to Wood's analysis.

His conclusion: slumping token prices plus Chinese models capturing market share make the projected US capex unsustainable.

The debt question

Wood has made the financing point before. In a Sept. 17 interview he said: "The key difference now is that this investment is increasingly being financed through debt rather than cash."

He added: "My base case remains that they will not generate returns sufficient to justify those investments. Ultimately there will be significant capital destruction."

His trigger is credit. "The moment credit markets effectively withdraw funding, the entire AI trade could unravel," he said. He gave no timetable and no specific market event.

Wood said in September that semiconductor companies could keep benefiting as long as investors and lenders remain willing to finance AI capex. That support would weaken if credit markets pull back or if infrastructure spending yields returns below expectations.

What the earnings show right now

The near-term numbers are strong, and Wood's own reports say so. In his Oct. 1 report, citing LSEG I/B/E/S data as of Sept. 25, S&P 500 companies were forecast to post 30.2% year-on-year third-quarter earnings growth, against 16.1% forecast a year earlier.

Information technology has the second-highest forecast growth among sectors at 65.1%, up from 22.3% last October. Semiconductors lead at 131.7%, and semiconductor materials and equipment follow at 68.9%.

Wood has credited that earnings momentum, driven significantly by the AI capex cycle, as the main reason US equities have withstood higher rates and geopolitical tension. He also noted that US stocks have historically tended to underperform ahead of midterm elections, a pattern that has not shown up this year because of the earnings growth.

The hyperscalers' own guidance, with capex projected to climb from $695 billion in 2026 to $870 billion in 2027, shows they are still committing capital at an accelerating pace.

Rates are the pressure point

The backdrop is tightening. The 10-year Treasury yield rose to 5.34% and the 30-year touched 5.69%, both the highest since 2002, according to Wood's newsletter last week. He pointed to a more hawkish Federal Reserve and said higher yields could put greater pressure on equity valuations.

That bears directly on his debt argument. The more borrowing costs rise, the more expensive it becomes to fund data centers on credit.

What to watch

Wood's call rests on two variables: whether lenders keep funding the buildout, and whether revenue growth catches up to the $695 billion now being spent. Neither has a date attached. The next test is the third-quarter earnings season, where the forecast 65.1% IT profit growth will be measured against actual results.